US7814564B2 - Method for fingerprinting multimedia content - Google Patents
Method for fingerprinting multimedia content Download PDFInfo
- Publication number
- US7814564B2 US7814564B2 US11/326,313 US32631306A US7814564B2 US 7814564 B2 US7814564 B2 US 7814564B2 US 32631306 A US32631306 A US 32631306A US 7814564 B2 US7814564 B2 US 7814564B2
- Authority
- US
- United States
- Prior art keywords
- group
- fingerprint
- codeword
- identifying
- participant
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related, expires
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012360 testing method Methods 0.000 claims abstract description 58
- 230000000875 corresponding effect Effects 0.000 claims description 72
- 238000004519 manufacturing process Methods 0.000 claims description 21
- 238000009826 distribution Methods 0.000 claims description 19
- 230000002596 correlated effect Effects 0.000 claims description 10
- 230000008901 benefit Effects 0.000 abstract description 9
- 238000001514 detection method Methods 0.000 description 60
- 238000012935 Averaging Methods 0.000 description 32
- 230000008569 process Effects 0.000 description 23
- 230000006872 improvement Effects 0.000 description 14
- 238000001228 spectrum Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 8
- 230000007480 spreading Effects 0.000 description 8
- 238000003892 spreading Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 5
- 239000000654 additive Substances 0.000 description 4
- 230000000996 additive effect Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000005192 partition Methods 0.000 description 3
- 238000007476 Maximum Likelihood Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000009432 framing Methods 0.000 description 2
- 230000008450 motivation Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000003467 diminishing effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/10—Protecting distributed programs or content, e.g. vending or licensing of copyrighted material ; Digital rights management [DRM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/10—Protecting distributed programs or content, e.g. vending or licensing of copyrighted material ; Digital rights management [DRM]
- G06F21/16—Program or content traceability, e.g. by watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/0028—Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0063—Image watermarking in relation to collusion attacks, e.g. collusion attack resistant
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0065—Extraction of an embedded watermark; Reliable detection
Definitions
- the invention described herein is directed to detecting the unauthorized copying of multimedia content by recognition of a previously embedded fingerprint extracted from the digital data comprising the content. More specifically, the invention is directed to exploiting the interaction between the structure of a fingerprint code and the embedding of such code into multimedia content so that the embedded code can be used to identify one of a plurality of colluders of attempting to degrade the code's usability.
- a fingerprinting system is considered collusion resistant when there is a high probability that one or more of the colluders can be positively identified from a suspicious copy of the fingerprinted multimedia content.
- a growing number of collusion-resistant fingerprinting techniques have been recently implemented on multimedia. The most used methods fall mainly into one of two categories according to whether explicit discrete coding is involved.
- a typical example of non-coded fingerprinting is orthogonal fingerprinting, which assigns each user a spread spectrum sequence as a fingerprint. Each fingerprint sequence is typically orthogonal to the fingerprints of other users.
- the resistance to collusion by orthogonal fingerprinting is often improved by grouping together fingerprint sequences according to characteristics of users who are likely to collude together.
- non-coded fingerprinting is easy to implement, the number of basis sequences required for its implementation and the computational complexity of detection increase linearly with the number of users assigned fingerprints.
- Coded fingerprints take advantage of code modulation spread spectrum techniques and have been applied in the past to generic data, such as executable software programs and bitstreams.
- Prior art coding applications operated under a “marking assumption”, which was introduced by Boneh and Shaw in “Collusion-secure Fingerprinting for Digital Data”, IEEE Trans. on Information Theory , 44(5) pp. 1897-1905, 1998.
- the marking assumption assumes that colluders change only those fingerprint symbols in the colluders' corresponding codes that have different values. It is further assumed that the colluders assemble pieces of their codes that are different from the codes of other colluders in order to generate a new untraceable version. Numerous fingerprinting schemes have been developed base on the marking assumption to expand upon Boneh and Shaw's framework.
- the “traceability (TA) code” may be constructed from structures resembling error correcting codes (ECC) and have been widely studied. Efficient decoding exists for some ECC, which makes such codes attractive for coded fingerprinting applications.
- ECC based fingerprinting the coded fingerprinting scheme constructed under ECC principles will be referred to as “ECC based fingerprinting”.
- the ECC based fingerprinting system includes an ECC based coding layer 100 and a spread spectrum based embedding layer 190 .
- the layers respectively occupy the fingerprinting stage 105 , where the fingerprints are applied to the multimedia content 135 , and a detection stage 195 , where the fingerprint codes are acquired from a test copy 150 of the content.
- the two stages are coupled by a distribution channel 145 , which may be a telecommunications channel or may be a market channel through which hard storage copy of the content, such as by a compact disk (CD) or digital versatile disk (DVD), is physically carried.
- a distribution channel 145 which may be a telecommunications channel or may be a market channel through which hard storage copy of the content, such as by a compact disk (CD) or digital versatile disk (DVD), is physically carried.
- One or more attackers 140 obtain the content somewhere along the distribution channel 145 to carry out the attack.
- each authorized user of the content is assigned a codeword, which is formed from ECC symbols selected from an alphabet of size q and having a large minimum distance.
- the encoder 110 partitions the content into a plurality of non-overlapped segments and encodes one symbol of the codeword in each segment.
- the partition can be made spatially into blocks for image, or temporally into frames for video and audio.
- the symbol is encoded as one of q mutually orthogonal spread spectrum sequences with identical energy via modulator 120 .
- Embedder 130 adds the sequence corresponding to the symbol value to the host signal at the appropriate segment.
- the embedder 130 may add the sequences in accordance with perceptual scaling known in the art so that the fingerprint will be unnoticed during presentation of the multimedia content.
- the fingerprinted multimedia content 135 is presented to the distribution channel 145 where it may be subject to an attack by one or more attackers 140 .
- Suspicious multimedia content 150 is the presented to an extractor 160 which retrieves the spread spectrum sequence from the content.
- the sequence is demodulated by demodulator 170 to retrieve the code and decoder 180 returns the user ID corresponding to the assigned fingerprint code.
- colluders contribute their assigned copies of the content segment by segment (or equivalently, symbol by symbol at the code level) with approximately equal share.
- a codeword for a user 1 was originally taken from codebook 210 and an ECC based codeword 220 was built thereon.
- a codeword 230 for user 2 was similarly constructed. Each colluding user 220 , 230 then provides roughly an equal number of segments containing their corresponding symbols to a colluded version 240 . Further distortion may be applied on the colluded signal, which is modeled as additive white Gaussian noise 250 .
- the colluded codeword 260 is first extracted by demodulating symbols from each segment. The colluder is then identified by, for example, matching the extracted codeword with an entry in the codebook.
- collusion attacks In another well-known type of collusion of collusion attack, the participants linearly combine their copies of the multimedia content to produce a colluded version. This generally reduces the energy in each contributed fingerprint until it is no longer discernible. Such collusion attacks are referred to herein as “averaging collusion” attacks.
- FIG. 3A and FIG. 3B An analytical approximation of the probability of detecting a single colluder, P d , for the ECC based fingerprinting under interleaving and averaging collusion is graphically illustrated in FIG. 3A and FIG. 3B , respectively.
- the Watermark-to-Noise-Ratio (WNR) ranges from 0 dB to ⁇ 20 dB, which includes the scenarios of severe distortion to mild distortion.
- the orthogonal fingerprinting case is shown in FIG. 3C and FIG. 3D for interleaving collusion and averaging collusion, respectively. Comparing the results of FIGS. 3B and 3D show that under averaging collusion, the orthogonal fingerprinting and ECC based fingerprinting constructed above have similar performance.
- the detection probability of the ECC based fingerprinting drops sharply when more than 6-7 colluders are involved in creating an interleaved copy, even when WNR is high.
- interleaving collusion is an effective strategy to circumvent the protection.
- a method for identifying a participant in production of an unauthorized version of multimedia content through a digital fingerprint embedded in digital data thereof.
- a codeword is assigned to each of a plurality of authorized users of the digital data.
- the codeword includes a plurality of symbols selected from a predetermined alphabet of symbols.
- the digital data is divided into a plurality of host data segments equal in number to the plurality of symbols forming the codeword.
- a fingerprint signal is encoded as a concatenation of digital code sequences corresponding to each symbol of the codeword.
- the fingerprint signal includes a plurality of fingerprint segments equal in number to the plurality of host data segments, whereby each of the fingerprint segments contains a corresponding one of the code sequences.
- the fingerprint segments are each segmented into a plurality of subsegments, which are then permuted and embedded into the digital data.
- a test fingerprint signal is subsequently extracted from a copy of the digital data and is segmented into the plurality of subsegments.
- the subsegments of the test fingerprint signal are inversely permuted with respect to the permutation of the subsegments of the fingerprint signal and the participant is identified as one of the authorized users assigned a codeword determined from the inversely permuted test fingerprint signal.
- a participant in production of an unauthorized version of multimedia content is identified by a method that first forms a plurality of user codewords each corresponding to a concatenation of coding sequences.
- the user codewords are grouped into a plurality of groups such that the concatenation of coding sequences corresponding to each of the user codewords in a corresponding one of the groups is orthogonal to remaining ones of such concatenated coding sequences in the group.
- the user codewords are assigned to a corresponding one of a plurality of authorized users such that at least one characteristic of the user is common to other users assigned a user codeword from that group.
- the digital data is divided into a plurality of host data segments equal in number to the plurality of coding sequences.
- a fingerprint signal is encoded as the concatenation of code sequences corresponding to each of the user codewords.
- the fingerprint signal includes a plurality of fingerprint segments equal in number to the plurality of host data segments and each of the fingerprint segments contains a corresponding one of the code sequences.
- Each of the fingerprint segments are segmented into a plurality of subsegments and are permuted.
- the permuted fingerprint signal is embedded into the digital data.
- a test fingerprint signal is extracted from a copy of the digital data and is segmented into the plurality of subsegments.
- the test fingerprint signal is inversely permuted with respect to the permutation of the subsegments of the fingerprint signal and the participant is identified as one of the authorized users assigned a codeword determined from the inversely permuted test fingerprint signal.
- each of a plurality of authorized users of the digital data is assigned a corresponding codeword including a plurality of symbols selected from a predetermined alphabet of symbols.
- Each symbol of the alphabet is encoded as a corresponding digital code sequence.
- the digital code sequences are segmented into a plurality of subsequences and the digital data of the multimedia content is segmented into a plurality of host data segments equal in number to the plurality of symbols forming said codeword, where each of the host data segments designated with a corresponding symbol of the codeword.
- Each of the host data segments is then segmented into a plurality of subsegments and each of the subsegments is designated with a portion of the symbol designated to the corresponding host data segment from which it was divided.
- a sub-segmented codeword is formed from the symbols designated to each of the subsegments, where the sub-segmented codeword is ordered identically to the order of the symbol portions of the designated subsegments.
- the sub-segmented codeword is then permuted.
- Each of the subsequences is embedded into a corresponding copy of each of the plurality of subsegments to produce a plurality of embedded subsegments equal in number to the number of symbols in the alphabet multiplied by the number of embedded subsegments.
- the embedded subsegments are concatenated in accordance with the permuted sub-segmented codeword to form a data signal and a test fingerprint signal is extracted therefrom such that the test fingerprint signal includes a plurality of subsegments equal in number to the number of subsegments in the sub-segmented codeword.
- the subsegments of the test fingerprint signal are inversely permuted with respect to the permutation of the subsegments of the sub-segmented codeword and the participant is identified as one of the authorized users assigned a codeword determined from the inversely permuted test fingerprint signal.
- FIG. 1 is a schematic block diagram of an ECC based multimedia fingerprinting system
- FIG. 2 is a diagram of an example of interleaving collusion
- FIG. 3A-B are graphs of the probability of detecting of a single participant in interleaving and averaging collusion on an ECC based fingerprinting system
- FIG. 3C-D are graphs of the probability of detecting of a single participant in interleaving and averaging collusion on an orthogonal fingerprinting system
- FIG. 4 is a diagram of permuted segment embedding (PSE) of ECC based fingerprint codes in accordance with the present invention
- FIG. 5 is a graph of probability of detecting of a single participant in interleaving collusion on a PSE fingerprinting system of the present invention as measured against the number of subsegments implemented;
- FIG. 6A is a graph of the probability of detecting of a single participant in segment interleaving collusion on a PSE fingerprinting system of the present invention
- FIG. 6B is a graph of the probability of detecting a single participant in subsegment interleaving collusion on a PSE fingerprinting system of the present invention.
- FIG. 7 is a block diagram illustrating fundamental aspects of group coding of fingerprints in accordance with the present invention.
- FIG. 8 is a flow diagram of fundamental method steps for partitioning a codebook into groups in accordance with the present invention.
- FIG. 9A is a diagram of group based joint coding and embedding (GRACE) in accordance with the present invention.
- FIG. 9B is a diagram of group coding by appending the group sub-codeword to the user sub-codeword in accordance with aspects of the present invention.
- FIGS. 10A-F are graphs depicting the effectiveness of the group coding implemented in accordance with the present invention.
- FIG. 11 is a block schematic diagram of a combined PSE and GRACE system in accordance with the present invention.
- FIG. 12A-F are graphs of the probability of detecting of a single participant in collusion on a combined PSE-GRACE system of the present invention.
- the present invention takes advantage of existing code construction techniques and adds additional measures that will be referred to herein as “permuted subsegment embedding” or PSE.
- PSE permuted subsegment embedding
- the function sym(j,k) is used to retrieve the symbol from the k th segment of the j th user's codeword, and u sym(j,k) is the spread spectrum sequence corresponding to the symbol value.
- colluders may manipulate fingerprinted multimedia in the signal domain, incurring a variety of code-domain changes that are inconsistent with the marking assumption. These signal domain attacks are said to occur on the embedding layer, since they manipulate the features of the signal without directly accessing the fingerprint code.
- a simple yet effective way alter the underlying fingerprint code is to average the corresponding signal components or features from multiple copies. This “averaging collusion” can be modeled as:
- s j represents the fingerprint sequence for user j
- S c the colluder set
- c the total number of colluders
- x the host signal
- d the additive noise term.
- s j represents the fingerprint sequence for user j
- S c the colluder set
- c the total number of colluders
- x the host signal
- d the additive noise term.
- the detection stage it is desired to identify one of the colluders with high probability, i.e., a high P d , the probability of identifying one colluder participating in the attack.
- certain applications of ECC based fingerprinting first determine which symbol is present in each segment by using a correlation detector commonly used in spread spectrum applications. For example, if the original multimedia signal, or host signal, is available to the detection stage, the suspicious copy may be registered with the host signal and the host data subsequently subtracted from that of the suspicious copy to obtain a test signal. Then, for each segment of the test signal, a maximum correlation detector is used to identify the symbol. This can be done by correlating each of the q spreading sequences with the test signal and the sequence resulting in the highest correlation statistic indicates the corresponding symbol.
- the detection statistic for the k th segment may be,
- z k and x k represent the k th segment of the colluded signal and that of the original signal, respectively.
- the extracted symbol from the k th segment is
- a soft-decision strategy may be employed to provide the correlation results of Equation (3) without having to determine each symbol value at every segment. The correlation is performed on all segments together to arrive at
- L is the code length
- N u is the total number of users.
- the function sym(j, k) is used to retrieve the symbol from the k th segment of the j th user's codeword. This approach corresponds to a matched filter detector that correlates the entire test signal with each user's fingerprint sequence s j by
- ⁇ s ⁇ ⁇ s j ⁇ for all j based upon the equal energy constraint applied during fingerprinting.
- the latter scheme takes advantage of the soft information on the symbol level and provides a better collusion identification performance.
- the non-coded orthogonal fingerprinting can be viewed as a special case of the alphabet size q equal to the total number of users N u and a codeword length of unity.
- the detection for orthogonal fingerprinting is done by first correlating the test signal with each user's sequence and then identifying the user with the highest correlation statistic as the colluder.
- the fingerprint code design establishes a minimum distance between codewords to be large enough so as to determine the code regardless of collusion attack technique.
- the minimum distance requirement ensures that the best match with a colluded codeword, often referred in the art as the “descendant”, comes from one of the true colluders.
- ⁇ ⁇ Q L be a code over an alphabet Q with length L and total number of codeword N u .
- a c-TA code can be constructed using an ECC if its minimum distance D satisfies
- c-TA code may be extended to account for both types of errors, i.e., tolerating L e erasures and L FA total errors and defining such c-TA code as c-TA q (L, N u ,L e L FA ). The minimum distance required then becomes
- ECC based fingerprint code prefers an ECC coding structure having larger minimum distance to identify a colluder from a larger set of participant colluders.
- the well-known Reed-Solomon code has the minimum distance that achieves the well-known Singleton bound and is used in existing coded fingerprinting works.
- Reed-Solomon codes may implement c-TA codes that satisfy the above conditions.
- a q-ary Reed-Solomon code of length L is used to construct a c-TA code with parameters satisfying
- the detection of the orthogonal fingerprinting may be achieved by correlating the test signal with each user's fingerprint sequence. This takes N u N multiplications plus N u (N ⁇ 1) summations, or a total of O(N u N) operations. Then, N u ⁇ 1 comparisons must be made to find the codeword corresponding to the highest correlation to identify at least one of the colluders.
- the computational complexity becomes O(qN).
- N u q t , where t is the dimension of the code.
- the computational complexity of ECC based fingerprinting for detection becomes O(t ⁇ square root over (N u ) ⁇ N).
- the significant improvement in the demodulation process is a big advantage of ECC based fingerprinting over the orthogonal fingerprinting.
- q orthogonal sequences of length NIL are used in ECC based system, while the orthogonal system requires N u mutually orthogonal sequences of length N. This suggests that the ECC based system provides advantageous compactness in representing users and consumes fewer resources in terms of the orthogonal sequences. This implies a much more efficient generation of fingerprint sequences in the embedding and detecting stages.
- the ECC based fingerprinting provides a potential support to efficient distribution of the fingerprinted signal.
- each segment is marked by one of only q possible symbols.
- q versions of each segment, each corresponding to a segment marked by each of the q symbols, is constructed in advance.
- the fingerprinted copy for each user is formed by concatenating the corresponding segments according to his/her codeword.
- the content owner or trusted distributor multicasts q versions encrypted with q different message keys to all the subscribers. After verification, the distributor delivers the message keys to corresponding users who can then decrypt their fingerprinted signal. If B denotes the bandwidth requirement of sending only one copy, q copies would require bandwidth of qB and a bandwidth of t ⁇ square root over (N u ) ⁇ B if a Reed-Solomon code is employed.
- a correlation-based statistic T N may be defined as
- s j is the fingerprint sequence for userj
- z is a colluded copy of the fingerprinted signal
- x is the original signal.
- the statistics for collusion participants, T 1 and innocent users, T 2 may be defined as,
- S c is the colluder set.
- T 1 and T 2 are approximated to be independent Gaussian distributed. Examination of the distribution of the correlations between each fingerprint sequence and the test sequence under interleaving collusion shows that the mean and variance of T 1 and T 2 may be expressed as:
- system performance is determined by a pairwise correlation denoted as ⁇ .
- ⁇ a pairwise correlation
- the first c users are assumed to perform the averaging collusion.
- the vector of detection statistics T N 's follows a N u -dimensional Gaussian distribution:
- the pairwise correlation of ECC based fingerprinting can be determined by examining the code construction. Codes with a larger minimum distance have a smaller upper bound on the correlation and thus are more preferable. Under the code construction with small minimum distance, the largest pairwise correlation between the fingerprinting sequences ⁇ o , which corresponds to the codewords with minimum distance, will be close to 0.
- a fingerprint signal 410 , 420 is generated by concatenating the spreading sequences corresponding to the symbols in a user's codeword.
- the codewords, formed in accordance with an ECC based fingerprinting system are maintained in a codebook 405 .
- Each original segment of the fingerprint signal is then divided into ⁇ subsegments, as shown at 422 and 424 , giving a total of ⁇ L subsegments.
- each of fingerprint sequences 410 and 420 are randomly permuted, as illustrated at 425 , to obtain the corresponding final fingerprint signals 430 and 435 , respectively.
- the permutation is randomized in accordance with a secret key. If interleaving collusion were then to proceed as previously described, the resultant descendant fingerprint would be that shown at 440 .
- the extracted fingerprint signal 440 is inversely permuted, as shown at 445 .
- the fingerprint signal 445 is then presented to a correlation detector, as shown at 450 , where it is correlated with the original fingerprints 422 and 424 , to produce a detection statistic 455 for each user.
- the detection statistic T N for the PSE system under interleaving collusion is modeled as an N u -dimensional Gaussian distribution:
- each colluded segment after interleaving collusion most likely contains subsegments from multiple users, which may be observed by correlation-based detectors (including both hard and soft detection at the symbol level). This is similar result to that brought about by averaging collusion. Since averaging collusion is far less effective from the colluder's point of view, the permuted subsegment embedding can greatly improve the collusion resistance of ECC based fingerprinting under interleaving collusion. Even if the colluders know the actual size of a segment or a subsegment, the permutation being unknown to them prevents them from creating a colluded signal with the equivalent effect of symbol interleaving in the code domain.
- the parameter ⁇ controls the “approximation” level of the effect of interleaving collusion to that of averaging collusion. Larger ⁇ provides a finer granularity in permutation. Thus, each segment may contain subsegments from more colluders, leading to higher similarity to averaging collusion detection and better collusion resistance.
- FIG. 6A-6B there are shown graphs illustrating the effectiveness of certain aspects of the invention.
- FIG. 6A shows colluder detection probabilities for segment-based interleaving collusion.
- the detection probability of the present invention is substantially improved over the original ECC based fingerprinting system when both processes are subjected to the interleaving collusion shown in FIG. 3A .
- the difference in performance between the present invention and that of the orthogonal fingerprinting is very small as shown by comparing FIG. 6A with FIG. 3C .
- FIG. 6B illustrates the detection probability for subsegment based interleaving collusion.
- 6A-6B show that the detection performance under interleaving collusion using a subsegment as a unit and using a segment as a unit are similar and give a high detection probability for up to two dozen colluders at moderate to high WNR. Since the permuted subsegment embedding does not affect the performance of the system under averaging collusion, the P d under averaging collusion remains unchanged. Overall, the PSE process of the present invention, which jointly considers coding and embedding layers, has effectively improved the collusion resistance.
- the detection in the exemplary PSE fingerprinting includes three steps: inverse permutation, demodulation by correlation and decoding to a certain colluder.
- the computational complexity of the inverse permutation is O( ⁇ L) and, as previously discussed, the demodulation and decoding steps require at most O(qN)+O(N u ) computations.
- PSE fingerprinting requires O( ⁇ L)+O(qN)+O(N u ).
- the largest possible value of ⁇ L is N, and usually N u ⁇ N.
- the demodulation step still dominates the overall complexity. Therefore, the overall computational complexity remains at O(qN).
- the efficient distribution may, in certain embodiments of the invention be achieved by dividing each sequence corresponding to every symbol in the alphabet into the same number ⁇ of subsequences.
- Each of the subsequences may be stored or transmitted to another location where fingerprinted versions of the multimedia content may be assembled.
- the subsequences are embedded into copies of each of the subsegments of the host signal to produce q different versions of the subsegment.
- the embedded subsegments are preformed to minimize impact on bandwidth, as previously discussed.
- the subsegments as embedded with the subsequences may be assembled in accordance with a permuted codeword, which may be inversely permuted at various points along the distribution channel when it is suspected that one or more embedded subsegments have fallen prey to attackers.
- a permuted codeword which may be inversely permuted at various points along the distribution channel when it is suspected that one or more embedded subsegments have fallen prey to attackers.
- the fingerprint signal may be processed as described above to reveal the identity of the attacker.
- the beneficial features of the permuted subsegment embedding of the present invention is the significant improvement in collusion resistance over ECC based fingerprinting with only a marginal increase in computation cost.
- the PSE fingerprinting may take advantage of the efficient distribution by pre-forming the subsegments.
- adjustments to user capacity can be made by modifying the dimension of the base ECC, e.g. the t of Reed-Solomon code.
- GRACE Group Based Joint Coding and Embedding
- fingerprint codewords may be placed into groups in accordance with common characteristics of authorized users based on the prior knowledge or assumptions on expected collusion patterns. Symbols representing the group information are added to the fingerprint, where the added information will be referred to herein as a “group sub-codeword”. In embodiments which include grouping, the codeword that is assigned to each user will be referred to as a “user sub-codeword”.
- a fingerprint codebook C is formed from error correcting codes over an alphabet of size q, such as the ECC based codes previously described.
- the code length is L
- the minimum distance is D, which is typically less than L.
- the codebook 715 is arranged into groups 725 to form a grouped codebook G, illustrated at 720 .
- Each user sub-codeword 730 a - 730 x is grouped so that the codewords of any particular group 725 are orthogonal to each other.
- users within a group are assigned codewords having distinct values at each symbol position and the group sub-codewords 735 a - 735 g may be subsequently assigned to the group attribute.
- the code distance within a group equals the codeword length L.
- FIG. 8 An exemplary process for grouping codewords orthogonally is illustrated in FIG. 8 .
- the process continues at block 820 , where an arbitrary codeword c ⁇ C is selected as the first element for group i.
- the code c is moved from C to group i.
- a codeword c ⁇ C is selected and it is determined at block 830 if the codeword is orthogonal to all the existing codewords in G(i).
- the process flow is transferred to block 835 , where the codeword c is moved from C to the group i, G(i), a step that is skipped if c is not orthogonal to other codewords c in G(i).
- the process continues at block 840 , where it is determined if the codebook C is empty. If not, flow is transferred to block 850 , where it is determined if all c ⁇ C have been checked for orthogonality. If all codewords have been checked, the process continues at block 845 , where the index i is incremented and the next group G(i) is set to the empty set. The process then continues at block 820 so as to build the next group. If it is determined at block 850 that more codewords need to be checked, flow is transferred back to block 825 , where the next codeword is retrieved. Once it is determined at block 840 that the codebook C is empty, the process ends at block 860 .
- the group sub-codewords themselves are designed to be orthogonal to one another.
- the group sub-codeword is a single symbol and g symbols would represent g groups. For each group, the symbol representing the group is repeated L times to form a group codeword, thereby ensuring that the symbol representing the group sub-codeword is embedded into each segment of the host signal.
- both the group sub-codeword and user sub-codeword are embedded into the host signal by mapping them to spreading sequences and then combining them with the host signal.
- the g sequences ⁇ a i ,i 1 . . .
- a higher ⁇ provides a more accurate detection of group information, but reduces the accuracy of user sub-codeword detection, which may lead to low accuracy of colluder detection, and vice versa. Since L segments are available to collect the energy for group detection and since collusion usually happens among a small number of groups, small ⁇ may be chosen to guarantee the user codeword detection with an acceptable group detection accuracy.
- the GRACE fingerprinting scheme embeds the spreading sequence of the group sub-codeword over that of the user sub-codeword so that the group information is spread throughout the signal.
- FIG. 9A Such process is illustrated in FIG. 9A , where a user sub-codeword 917 may be divided into subsegments, as previously described.
- a group sub-codeword 927 is shown as being composed of a number of different symbols, however the group sub-codeword may be composed of repeated version of the same symbol, as previously described.
- the user sub-codeword 917 and the group sub-codeword may be divided into a plurality of subsegments 922 .
- the user sub-codeword 917 and the group sub-codeword 927 are randomly permuted, as illustrated at 932 .
- a permuted fingerprint sequence is produced according to the permuted subsegments 937 and the group fingerprint sequence 937 and 942 .
- the fingerprint sequences 937 , 942 are then presented to the embedding layer 947 , where they are embedded into the host signal to produce the fingerprinted signal.
- the group information is embedded by appending the spreading sequence of group sub-codeword to that of user sub-codeword, as shown in FIG. 9B .
- the total sample budget of the host signal 910 for purposes of fingerprinting is N.
- the length in symbols of the user codeword 920 is L and the length in symbols of the group sub-codeword 930 is L g .
- the segment length is adjusted to accommodate appending the group sub-codeword to the user sub-codeword, as illustrated at 940 .
- the length of the group sub-codeword, L g regulates the relative energy between the group sub-codeword 930 and user sub-codeword 920 .
- the group symbols may be extracted from their corresponding positions through a non-blind correlation detector, where the correlation is done over all the segments containing the group sub-codeword.
- the groups are considered guilty of collusion if the detection statistics of the test signal are greater than a certain threshold.
- a set of guilty groups may be determined by the union of the detected guilty groups.
- each group has different group sub-codeword from the others, and within one group users have different user sub-codeword.
- the fingerprint sequence for each user is different from any other user.
- colluders cannot figure out the grouping information through comparing their copies. And no matter which segment the colluder contributes, he/she always contributes both the group information and user information. Such a group-framing attack is unlikely to succeed under the present invention.
- the joint coding and embedding of GRACE provides a more secure and effective way to incorporate the group information.
- certain embodiments of the invention implement a two-level detection scheme.
- a correlation detector is used to examine the group information in the colluded signal to identify the groups to which the colluders belong.
- the identified guilty groups are used to narrow down the set of fingerprints on which ECC decoding need be applied.
- group information is extracted from the test signal y using a non-blind correlation detector, the detection statistic with respect to group i is
- the user sub-codeword is extracted from each segment of the test signal z through, for example, the maximum correlation detector previously described, with each of the q spreading sequences u j .
- An ECC based decoder may then be used to decode the codeword and the codebook is examined to find the codeword within the guilty groups closest to the extracted codeword.
- the detector declares the user corresponding to this codeword as the colluder.
- the soft detection technique which correlates the test signal with each user's fingerprint sequence, may be employed to identify the sequence with the highest correlation statistic.
- L g 5 was chosen for the embedding by appending strategy so that the relative energy between user sub-codeword and group sub-codeword is the same as that of GRACE. Repeated symbols were used as the group sub-codewords, such as described above. Independent identically distributed Gaussian signals with 3 ⁇ 10 4 signal samples were used to emulate the host signal.
- interleaving collusion and averaging collusion were applied to three exemplary systems—ECC based fingerprinting, GRACE fingerprinting and the Group based ECC fingerprinting with appending embedding.
- the chosen grouping correctly reflects the collusion pattern of the colluder's distribution so that all the colluders come from a small number of groups.
- all colluders are from two out of 32 groups, and are randomly distributed between these two groups.
- the results of P d under interleaving collusion and averaging collusion are shown in FIGS. 10A and 10B , respectively.
- the P d 's for the GRACE process and the appending embedding strategy are similar and have 10% improvement over ECC based fingerprinting.
- the improvement in P d can be as large as 70%.
- P d is required to be no less than a certain value, e.g. 0.98, the number of colluders the system can resist can be improved from 6 colluders (for ECC based fingerprinting) to 18 colluders (for the proposed GRACE).
- a certain value e.g. 0.98
- the number of colluders the system can resist can be improved from 6 colluders (for ECC based fingerprinting) to 18 colluders (for the proposed GRACE).
- a certain value e.g. 0.98
- the detector may lose the information of some guilty groups, which results in no performance improvement over ECC based fingerprinting.
- both group-based schemes have the same group information from the colluders, so their performances are similar. Due to the participation of multiple groups, the energy of the group information is reduced and extracted with low accuracy; leading to diminishing performance gain over ECC based fingerprinting.
- FIG. 11 An exemplary combined PSE-GRACE system is illustrated in FIG. 11 .
- permuted subsegment embedding is applied on the fingerprint sequence of both user sub-codeword and group sub-codeword fingerprint sequences.
- the user sub-codewords are encoded in encoder 1110 and the group sub-codewords are encoded in the GRACE encoder 1115 .
- the GRACE encoder 1115 provides a user sub-codeword on line 1117 a and a group sub-codeword on 1117 b to modulator 1120 , which produces a user fingerprint sequence on line 1123 a and a group fingerprint sequence on line 1123 b .
- the user fingerprint sequence is superposed on the group fingerprint sequence and the combination is permuted to provide a combined fingerprint sequence to embedder 1130 , wherein the host signal is fingerprinted.
- the fingerprinted multimedia 1135 is presented to the distribution channel 1140 , where it may undergo an attack 1143 .
- the suspect multimedia 1145 is presented to a two level detector 1185 , whereby group information extraction is followed by the soft detection of the colluder from within the identified groups, as described above.
- the combined fingerprint sequence is first extracted from the host signal by extractor 1150 .
- the fingerprint is presented to inverse permuter and separator 1160 , where user information is provided on line 1187 a and the group information on line 1187 b .
- the group sequence is demodulated at demodulator 1165 .
- the user fingerprint sequence is demodulated at 1175 and passed to decoder 1190 , where the most probable symbol is extracted from each segment.
- test signal is directly correlated with each user's fingerprint via correlator 1180 .
- the correlation statistic provided by correlator 1180 is examined and the user having the highest correlation is declared a participant in the collusion.
- the identity of the guilty user is provided at 1195 .
- the performance of the combined PSE-GRACE fingerprinting system is compared according to the same criteria used in the comparisons made with regard to FIGS. 10A-10F .
- the combination of PSE and GRACE achieves better results than each individual approach as is shown in FIGS. 12A-12F .
- the permuted subsegment embedding with GRACE improves the collusion resistance of the fingerprinting system by 40% more under interleaving collusion at high WNR.
- the grouping strategy boosts up the system performance of P d to nearly 1 for the value ranges of WNR and c examined with respect to FIGS. 10A-10B .
- the extra computation of the GRACE Fingerprinting detection comes from the detection of guilty groups which requires O(gN) computations, where g is the total number of groups. From previous discussions, it can be readily observed that the overall computational complexity for GRACE is O(qN)+O(gN). Since g is the group number and is usually much smaller than the total number of users, the computational complexity of GRACE remains on the same order as the standard ECC based system.
- the suspicious user set will be much smaller than that in non-grouped ECC based fingerprinting. This provides faster detection process during the colluder detection stage, resulting in a faster detection than the above analysis.
- GRACE fingerprinting is to use the group information to narrow the suspicious colluders down to a smaller group. Within each group the minimum distance between the users' codewords are larger than that of the whole user set so that the users' codewords are more separated and easier to detect. Following this idea, we can extend our GRACE fingerprinting to general multi-level GRACE to capture a more complicated collusion pattern.
- the codebook is partitioned or built with minimum distance D 0 into groups. Inside each group, the minimum distance D 1 is larger than D 0 . This partition is repeated for each group until the minimum distance equals the code length L or the structure of the group can capture the collusion pattern.
- the same strategy is used in a tree-based scheme to assign at each level an orthogonal sequence and embed them by proper scaling. On the detector side, the group information at each level is used one by one to narrow down the suspicious colluders to a smaller group. Then the colluder can be detected inside the extracted group as before.
Abstract
Description
y jk =u sym(j,k) +x k (1)
where xk is kth segment of host signal, and yjk is the kth fingerprinted segment for the jth user. The function sym(j,k) is used to retrieve the symbol from the kth segment of the jth user's codeword, and usym(j,k) is the spread spectrum sequence corresponding to the symbol value. The symbols are taken from a set of q mutually orthogonal spread spectrum sequences {ui:i=1, . . . , q} with identical energy ∥u∥2. The concatenation of all fingerprinted segments ultimately forms the fingerprinted signal, whereas the concatenation of the spread spectrum sequences uk, k=1, . . . , L, where L is the length of the codeword, is referred to as the fingerprint signal.
where sj represents the fingerprint sequence for user j, Sc the colluder set, c the total number of colluders, x the host signal, and d the additive noise term. Studies have shown that a number of nonlinear collusion processes can be approximated by an averaging process with the addition of a noise term. For simplicity, the assumption is generally made that the additional noise in both interleaving and averaging collusion follows i.i.d. Gaussian distribution. The effects of other distortions have been studied in watermarking applications, such as quantization/compression and geometric distortions. When the original host signal is available, the effects of many distortions attempting to dilute the fingerprint can be approximated quite well by additive noise.
where zk and xk represent the kth segment of the colluded signal and that of the original signal, respectively. The extracted symbol from the kth segment is
With the sequence of signals extracted from all segments using this maximum detector, a decoding procedure is applied to identify the colluder whose codeword has the most matched symbols with the extracted symbol sequence.
where L is the code length and Nu is the total number of users. The function sym(j, k) is used to retrieve the symbol from the kth segment of the jth user's codeword. This approach corresponds to a matched filter detector that correlates the entire test signal with each user's fingerprint sequence sj by
Here, ∥s∥=∥sj∥ for all j based upon the equal energy constraint applied during fingerprinting. The user who is assigned the fingerprint code having the highest correlation value TN(j), i.e. ĵ=arg maxj=1, 2, . . . , N
desc(C)={[x 1 . . . x L ]:x j ε{w j (i):1≦i≦c}, 1≦j≦L} (7)
If for any descendant [x1 . . . xL]εdesc(C), there is a viεC such that
|{j:x j =w j (i) }|>|{j:x j =S j}|, (8)
for any innocent user's codeword [s1 . . . sL]εΓ/C where the notation |•| is the cardinality operator, then Γ is called a c-traceability (c-TA) code and denoted as c-TAq(L,Nu) with q=|Q|.
where c is the colluder number and L is the code length. It is then said that Γ is a c-TAq(L, Nu) code.
where Le is the number of erasures.
which follows multivariate Gaussian distribution of Nu dimensions. Here, sj is the fingerprint sequence for userj, z is a colluded copy of the fingerprinted signal and x is the original signal. The statistics for collusion participants, T1 and innocent users, T2, may be defined as,
where Sc is the colluder set. For simplicity, T1 and T2 are approximated to be independent Gaussian distributed. Examination of the distribution of the correlations between each fingerprint sequence and the test sequence under interleaving collusion shows that the mean and variance of T1 and T2 may be expressed as:
where ∥u∥ is the strength of the spread spectrum sequence corresponding to each symbol in the alphabet of fingerprint codes, L is the length of the fingerprint code, D is the minimum distance between symbols in the alphabet, c is the number of colluders and σd 2 is the variance of the additive noise. Then,
P d =Pr(T 1 >T 2)=∫28 −∞ P(T 1 >t)ƒT
where ƒT
where 1k is an all-1 vector with dimension k-by-1, and Σ is an n-by-n matrix whose diagonal elements are 1's and off-diagonal elements are ρ's, σd 2 is the variance of the noise, m1 is the mean vector for colluders and m2 is the mean vector for innocent users. Given the same colluder number c and fingerprint strength ∥s∥, the mean correlation values for colluders and for innocents are separated more widely for a smaller ρ. Thus, absent any prior knowledge of the collusion pattern, a smaller ρ leads to a larger colluder detection probability Pd. Therefore, fingerprint sequences with a small pairwise correlation ρ are preferred in designing a fingerprinting system.
where Σ is as described above for averaging collusion with ρ=(L−D)/L. The variable β then serves as a parameter that controls the similarity of the PSE detection results to those of the averaging collusion detection. As will be shown, larger β provides a finer granularity in subsegment division and permutation.
s ijk=√{square root over (1−ρ)}u sym(j,k) +√{square root over (ρ)}a i, (18)
where function sym(kj) is used to retrieve the kth symbol in jth segment, ρ is used to adjust the relative energy between the group sub-codeword and user sub-codeword. This fingerprint signal is added to the jth segment of the host signal. A higher ρ provides a more accurate detection of group information, but reduces the accuracy of user sub-codeword detection, which may lead to low accuracy of colluder detection, and vice versa. Since L segments are available to collect the energy for group detection and since collusion usually happens among a small number of groups, small ρ may be chosen to guarantee the user codeword detection with an acceptable group detection accuracy.
where bi is the concatenation of the sequences representing group i's information from each segment. For example, when a single repeated symbol ai is embedded as the group sub-codeword, in each segment of group i, bi=[ai . . . ai]. The kth group is considered guilty for the test signal if TG(k)>h, where h is a predetermined threshold.
Claims (19)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/326,313 US7814564B2 (en) | 2005-01-07 | 2006-01-06 | Method for fingerprinting multimedia content |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US64228605P | 2005-01-07 | 2005-01-07 | |
US11/326,313 US7814564B2 (en) | 2005-01-07 | 2006-01-06 | Method for fingerprinting multimedia content |
Publications (2)
Publication Number | Publication Date |
---|---|
US20070180537A1 US20070180537A1 (en) | 2007-08-02 |
US7814564B2 true US7814564B2 (en) | 2010-10-12 |
Family
ID=38323715
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/326,313 Expired - Fee Related US7814564B2 (en) | 2005-01-07 | 2006-01-06 | Method for fingerprinting multimedia content |
Country Status (1)
Country | Link |
---|---|
US (1) | US7814564B2 (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070165853A1 (en) * | 2005-12-30 | 2007-07-19 | Hongxia Jin | Method for tracing traitor coalitions and preventing piracy of digital content in a broadcast encryption system |
US20080056497A1 (en) * | 2006-09-06 | 2008-03-06 | Korea Advanced Institute Of Science And Technology | Method of generating anti-collusion fingerprint codes using |
US20110182466A1 (en) * | 2008-09-26 | 2011-07-28 | Thomson Licensing A Corporation | Method for protecting digital content against minority collusion attacks |
US20120072593A1 (en) * | 2008-09-26 | 2012-03-22 | Ju-Yeob Kim | Multimedia content file management system for and method of using genetic information |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8205237B2 (en) | 2000-09-14 | 2012-06-19 | Cox Ingemar J | Identifying works, using a sub-linear time search, such as an approximate nearest neighbor search, for initiating a work-based action, such as an action on the internet |
US7814564B2 (en) * | 2005-01-07 | 2010-10-12 | University Of Maryland | Method for fingerprinting multimedia content |
US20070162761A1 (en) * | 2005-12-23 | 2007-07-12 | Davis Bruce L | Methods and Systems to Help Detect Identity Fraud |
US9654447B2 (en) * | 2006-08-29 | 2017-05-16 | Digimarc Corporation | Customized handling of copied content based on owner-specified similarity thresholds |
US8738749B2 (en) | 2006-08-29 | 2014-05-27 | Digimarc Corporation | Content monitoring and host compliance evaluation |
US8707459B2 (en) | 2007-01-19 | 2014-04-22 | Digimarc Corporation | Determination of originality of content |
US8010511B2 (en) | 2006-08-29 | 2011-08-30 | Attributor Corporation | Content monitoring and compliance enforcement |
US8301658B2 (en) * | 2006-11-03 | 2012-10-30 | Google Inc. | Site directed management of audio components of uploaded video files |
CN101730902A (en) * | 2007-05-03 | 2010-06-09 | 谷歌公司 | Monetization of digital content contributions |
US8094872B1 (en) | 2007-05-09 | 2012-01-10 | Google Inc. | Three-dimensional wavelet based video fingerprinting |
US8611422B1 (en) | 2007-06-19 | 2013-12-17 | Google Inc. | Endpoint based video fingerprinting |
US7975313B2 (en) * | 2007-08-14 | 2011-07-05 | International Business Machines Corporation | System and method for tracing Tardos fingerprint codes |
FR2925724A1 (en) * | 2007-12-24 | 2009-06-26 | France Telecom | METHOD FOR ANALYZING MULTIMEDIA CONTENT, COMPUTER PROGRAM PRODUCT AND CORRESPONDING ANALYZING DEVICE |
CN102227745B (en) * | 2008-09-26 | 2014-05-28 | 汤姆森特许公司 | Method for constructing inner codes for anti-collusion forensic code for watermarking digital content |
US8495733B1 (en) | 2009-03-25 | 2013-07-23 | Trend Micro Incorporated | Content fingerprinting using context offset sequences |
US9633014B2 (en) * | 2009-04-08 | 2017-04-25 | Google Inc. | Policy based video content syndication |
US8229488B2 (en) * | 2009-07-30 | 2012-07-24 | Sony Ericsson Mobile Communications Ab | Methods, apparatuses and computer programs for media content distribution |
US20110213720A1 (en) * | 2009-08-13 | 2011-09-01 | Google Inc. | Content Rights Management |
US8424091B1 (en) | 2010-01-12 | 2013-04-16 | Trend Micro Incorporated | Automatic local detection of computer security threats |
US9812138B1 (en) * | 2014-09-03 | 2017-11-07 | Amazon Technologies, Inc. | Proving file ownership |
CN110223213B (en) * | 2019-06-14 | 2022-10-04 | 兰州交通大学 | Vector space data digital fingerprint method for GD-PBIBD coding |
US20220207120A1 (en) * | 2020-12-30 | 2022-06-30 | Inka Entworks, Inc | Apparatus and method for embedding plurality of forensic marks |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6076165A (en) * | 1998-01-16 | 2000-06-13 | Sony Corporation | Method for authenticating digital recording devices |
US6104826A (en) * | 1997-02-19 | 2000-08-15 | Fujitsu Limited | Method of watermark-embedding/extracting identification information into/from picture data and apparatus thereof, and computer readable medium |
US6246349B1 (en) * | 1998-12-11 | 2001-06-12 | International Business Machines Corporation | Method and system for compressing a state table that allows use of the state table without full uncompression |
US20040010692A1 (en) * | 2002-07-09 | 2004-01-15 | Kaleidescape | Detecting collusion among multiple recipients of fingerprinted information |
US6691229B1 (en) * | 2000-03-06 | 2004-02-10 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for rendering unauthorized copies of digital content traceable to authorized copies |
US6754364B1 (en) * | 1999-10-28 | 2004-06-22 | Microsoft Corporation | Methods and systems for fingerprinting digital data |
US7068823B2 (en) * | 2003-06-23 | 2006-06-27 | Sony Pictures Entertainment Inc. | Fingerprinting of data |
US20070180537A1 (en) * | 2005-01-07 | 2007-08-02 | Shan He | Method for fingerprinting multimedia content |
-
2006
- 2006-01-06 US US11/326,313 patent/US7814564B2/en not_active Expired - Fee Related
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6104826A (en) * | 1997-02-19 | 2000-08-15 | Fujitsu Limited | Method of watermark-embedding/extracting identification information into/from picture data and apparatus thereof, and computer readable medium |
US6076165A (en) * | 1998-01-16 | 2000-06-13 | Sony Corporation | Method for authenticating digital recording devices |
US6246349B1 (en) * | 1998-12-11 | 2001-06-12 | International Business Machines Corporation | Method and system for compressing a state table that allows use of the state table without full uncompression |
US6754364B1 (en) * | 1999-10-28 | 2004-06-22 | Microsoft Corporation | Methods and systems for fingerprinting digital data |
US7177442B2 (en) * | 1999-10-28 | 2007-02-13 | Microsoft Corporation | Methods and systems for fingerprinting digital data |
US6691229B1 (en) * | 2000-03-06 | 2004-02-10 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for rendering unauthorized copies of digital content traceable to authorized copies |
US20040010692A1 (en) * | 2002-07-09 | 2004-01-15 | Kaleidescape | Detecting collusion among multiple recipients of fingerprinted information |
US7036024B2 (en) * | 2002-07-09 | 2006-04-25 | Kaleidescape, Inc. | Detecting collusion among multiple recipients of fingerprinted information |
US7068823B2 (en) * | 2003-06-23 | 2006-06-27 | Sony Pictures Entertainment Inc. | Fingerprinting of data |
US20070180537A1 (en) * | 2005-01-07 | 2007-08-02 | Shan He | Method for fingerprinting multimedia content |
Non-Patent Citations (2)
Title |
---|
Boneh, D., et al., "Collusion-Secure Fingerprinting for Digital Data", IEEE Transactions on Information Theory, vol. 44, No. 5, Sep. 1998, pp. 1897-1905. |
Safavi-Naini, R., et al., 18 page preprint of "Collusion Secure q-ary Fingerprinting for Perceptual Content", Security and Privacy in Digital Rights Management (SPDRM'01), 2002, pp. 57-75. |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070165853A1 (en) * | 2005-12-30 | 2007-07-19 | Hongxia Jin | Method for tracing traitor coalitions and preventing piracy of digital content in a broadcast encryption system |
US8176568B2 (en) * | 2005-12-30 | 2012-05-08 | International Business Machines Corporation | Tracing traitor coalitions and preventing piracy of digital content in a broadcast encryption system |
US8495755B2 (en) | 2005-12-30 | 2013-07-23 | International Business Machines Corporation | Tracing traitor coalitions and preventing piracy of digital content in a broadcast encryption system |
US20080056497A1 (en) * | 2006-09-06 | 2008-03-06 | Korea Advanced Institute Of Science And Technology | Method of generating anti-collusion fingerprint codes using |
US7953975B2 (en) * | 2006-09-06 | 2011-05-31 | Korea Advanced Institute Of Science And Technology | Method of generating anti-collusion fingerprint codes using (na,n2a−2,na−1,n,0,1) GD-PBIBD |
US20110182466A1 (en) * | 2008-09-26 | 2011-07-28 | Thomson Licensing A Corporation | Method for protecting digital content against minority collusion attacks |
US20120072593A1 (en) * | 2008-09-26 | 2012-03-22 | Ju-Yeob Kim | Multimedia content file management system for and method of using genetic information |
US8995706B2 (en) * | 2008-09-26 | 2015-03-31 | Thomson Licensing | Method for protecting digital content against minority collusion attacks |
Also Published As
Publication number | Publication date |
---|---|
US20070180537A1 (en) | 2007-08-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7814564B2 (en) | Method for fingerprinting multimedia content | |
He et al. | Joint coding and embedding techniques for multimediafingerprinting | |
Moulin et al. | Data-hiding codes | |
US8032754B2 (en) | Systems and methods for embedding media forensic identification markings | |
US7382905B2 (en) | Desynchronized fingerprinting method and system for digital multimedia data | |
He et al. | Collusion-resistant video fingerprinting for large user group | |
Zhao et al. | Fingerprint multicast in secure video streaming | |
Wu | Linear combination collusion attack and its application on an anti-collusion fingerprinting [image watermarking] | |
Xie et al. | On-off keying modulation and Tardos fingerprinting | |
Ji et al. | The curse of correlations for robust fingerprinting of relational databases | |
Zaidan et al. | Stego-image vs stego-analysis system | |
Chaabane et al. | A multimedia tracing traitors scheme using multi-level hierarchical structure for tardos fingerprint based audio watermarking | |
He et al. | Performance study of ECC-based collusion-resistant multimedia fingerprinting | |
US8516608B2 (en) | Method for decoding a probabilistic anti-collusion code | |
He et al. | Improving collusion resistance of error correcting code based multimedia fingerprinting | |
US8370958B2 (en) | Decoding method for a probabilistic anti-collusion code comprising the selection of the collusion strategy | |
Karthik et al. | Decryption key design for joint fingerprinting and decryption in the sign bit plane for multicast content protection. | |
Rezaei et al. | A novel steganography algorithm using edge detection and MPC algorithm | |
He et al. | Group-based joint coding and embedding technique for multimedia fingerprinting | |
Watters et al. | Visual detection of LSB-encoded natural image steganography | |
Wu et al. | Collusion-resistant multimedia fingerprinting: a unified framework | |
Li | Perceptual Multimedia Hashing | |
Li et al. | Multimedia Content Tracing Scheme Based on Tardos' Fingerprint Codes | |
Wang | Detection-and information-theoretic analysis of steganography and fingerprinting | |
Trappe et al. | Joint coding and embedding for collusion-resistant fingerprinting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: UNIVERSITY OF MARYLAND, MARYLAND Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HE, SHAN;WU, MIN;SIGNING DATES FROM 20060320 TO 20060405;REEL/FRAME:017628/0460 Owner name: UNIVERSITY OF MARYLAND, MARYLAND Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:HE, SHAN;WU, MIN;REEL/FRAME:017628/0460;SIGNING DATES FROM 20060320 TO 20060405 |
|
STCF | Information on status: patent grant |
Free format text: PATENTED CASE |
|
FEPP | Fee payment procedure |
Free format text: PATENT HOLDER CLAIMS MICRO ENTITY STATUS, ENTITY STATUS SET TO MICRO (ORIGINAL EVENT CODE: STOM); ENTITY STATUS OF PATENT OWNER: MICROENTITY |
|
FPAY | Fee payment |
Year of fee payment: 4 |
|
MAFP | Maintenance fee payment |
Free format text: PAYMENT OF MAINTENANCE FEE, 8TH YEAR, MICRO ENTITY (ORIGINAL EVENT CODE: M3552) Year of fee payment: 8 |
|
FEPP | Fee payment procedure |
Free format text: MAINTENANCE FEE REMINDER MAILED (ORIGINAL EVENT CODE: REM.); ENTITY STATUS OF PATENT OWNER: MICROENTITY |
|
LAPS | Lapse for failure to pay maintenance fees |
Free format text: PATENT EXPIRED FOR FAILURE TO PAY MAINTENANCE FEES (ORIGINAL EVENT CODE: EXP.); ENTITY STATUS OF PATENT OWNER: MICROENTITY |
|
STCH | Information on status: patent discontinuation |
Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362 |
|
FP | Lapsed due to failure to pay maintenance fee |
Effective date: 20221012 |